CN105409241A - Microphone calibration - Google Patents

Abstract

The disclosed apparatus, systems, and methods provide a calibration technique for calibrating a set of microphones. The disclosed calibration technique is configured to calibrate the microphones with respect to a reference microphone and can be used in actual operation rather than a testing environment. The disclosed calibration technique can estimate both the magnitude calibration factor for compensating magnitude sensitivity variations and the relative phase error for compensating phase delay variations. In addition, the disclosed calibration technique can be used even when multiple acoustic sources are present. The disclosed technique is particularly well suited to calibrating a set of microphones that are omnidirectional and sufficiently close to one another.

Description

Microphone calibration

The cross reference of related application

This application claims the rights and interests of comparatively early priority date that submit to July 26 in 2013, the U.S. Provisional Patent Application that is entitled as " APPARATUS; SYSTEMS; ANDMETHODSFORMICROPHONECALIBRATION " number 61/858750, it is incorporated to by reference at this clearly, and it is overall.

Background

Technical field

Disclosed device, system and method relate to the microphone be aligned in electronic system.

Background technology

Electronic equipment uses multiple microphone usually, to improve the quality of the acoustic intelligence of measurement, and extracts the information of regarding sound-source and/or surrounding environment.Such as, electronic equipment can use the signal detected by multiple microphone to be separated them with the source according to them, and this is commonly called blind source separating.As another example, electronic equipment can use the signal detected by multiple microphone to suppress the echo in detection signal or to cancel sound equipment echo from detection signal.

When processing the signal detected by multiple microphone, electronic equipment supposes that microphone has identical amplitude sensitivity and phase error usually.Unfortunately, microphone does not have identical amplitude sensitivity and phase error usually, even when using identical process to create microphone.Such technique change is more obvious in the microphone of the cheapness for consumer electronics product (such as, smart mobile phone).Because the value sensitivity of appropriate variance and/or phase error can cause the appreciable error of above-mentioned application, be necessary the device, the system and method that are provided in the art calibrating microphone.

Summary of the invention

In this application, provide for device, the system and method at electronic system alignment microphone.

Some embodiments comprise a kind of device.This device can comprise interface, is configured as receiving the first digitized signal stream and the second digitized signal stream, and wherein, described first digitized signal stream and the second digitized signal stream correspond to the voice signal of being caught by the first microphone and second microphone respectively.This device can also comprise the processor with described interface communication, is configured to run in the module stored in memory.This module can be configured to determine described first digitized signal stream the very first time-the second T/F of frequency representation and the second digitized signal stream represents, wherein, the very first time-frequency representation indicates the first digitized signal stream for the amplitude of the multiple frequencies at multiple time frame, and wherein said second T/F represents the size of instruction second digitized signal stream for multiple frequencies of multiple time frame; Determine the very first time-frequency representation and the second T/F represent at multiple time frame for the relation between more than first frequency; And based on the very first time-frequency representation and the second T/F represent, determines the relation for the amplitude calibration factor of more than first frequency between the first microphone and second microphone.

Some embodiments comprise a kind of method.The method can comprise: receive the first digitized signal stream and the second digitized signal stream by the data processing module being coupled to the first microphone and second microphone, wherein, described first digitized signal stream and the second digitized signal stream correspond to the voice signal of catching respectively by the first microphone and second microphone.The method can also comprise: by data processing module determine described first digitized signal stream the very first time-the second T/F of frequency representation and the second digitized signal stream represents, wherein, the very first time-frequency representation represents the size of the first digitized signal stream for multiple frequencies of multiple time frame, and wherein said second T/F expression second digitized signal stream is for the amplitude of multiple frequencies of multiple time frame.The method may further include: by the calibration module communicated with described data processing module determine described the very first time-frequency representation and the second T/F represent between in the relation of multiple time frame for multiple first frequency.The method can also comprise: based on the very first time-relation between frequency representation and the second T/F representation, determine the first microphone and the second microphone amplitude calibration factor for more than first frequency by calibration module.

Some embodiments comprise non-transitory computer-readable medium.Non-transitory computer-readable medium can comprise executable instruction, can operate to impel data processing equipment to pass through to be coupled to interface first digitized signal stream and the second digitized signal stream of the first microphone and second microphone, wherein said first digitized signal stream and the second digitized signal stream correspond to the voice signal of being caught by described first microphone and second microphone respectively.Described computer-readable medium can also comprise executable instruction, can operate to make data processing equipment, with determine described first digitized signal stream the very first time-the second T/F of frequency representation and the second digitized signal stream represents, wherein, the very first time-frequency representation indicates the first digitized signal stream for the size of multiple frequencies of multiple time frame, and wherein said second T/F represents the size of instruction second digitized signal stream for the multiple frequencies for multiple time frame.Described computer-readable medium can also comprise executable instruction, can operate to make data processing equipment determine described the very first time-between frequency representation and described second T/F representation for the relation of more than first frequency in multiple time frame, and based on the very first time-relation between frequency representation and the second T/F representation, determine the first microphone and the second microphone amplitude calibration factor for more than first frequency.

In certain embodiments, this device, method and/or non-transitory computer-readable medium can comprise module, step or executable instruction, for determine for the second T/F of more than first frequency represent with the very first time-frequency representation for the ratio of each multiple time frame, and determines the histogram of the ratio corresponding to more than first frequency.

In certain embodiments, this device, method and/or non-transitory computer-readable medium can comprise module, step or executable instruction, for determining amplitude correction factor based on the counting of ratio in histogram.

In certain embodiments, this device, method and/or non-transitory computer-readable medium can comprise module, step or executable instruction, for determining the multiple amplitude calibration factors corresponding to multiple frequency based on multiple histogram, wherein said multiple histogram corresponds respectively to multiple frequency; With the amplitude calibration factor that is level and smooth and at least two described multiple frequency dependence connection.

In certain embodiments, this device, method and/or non-transitory computer-readable medium can comprise module, step or executable instruction, for identifying the ratio in histogram with the highest counting.

In certain embodiments, device, method and/or non-transitory computer-readable medium can comprise module, step or executable instruction, for identify modeling correspond to described multiple time frame and more than first frequency described the very first time-frequency representation and the second T/F represent between the line of relation.

In certain embodiments, this device, method and/or non-transitory computer-readable medium can comprise module, step or executable instruction, for by more than first frequency the very first time-frequency representation is multiplied by the amplitude calibration factor of more than first frequency, to calibrate the first microphone relative to described second microphone.

In certain embodiments, this device, method and/or non-transitory computer-readable medium can comprise module, step or executable instruction, correspond to by the first additional character signal of the first microphone at the first digitized signal stream of the voice signal of very first time frame-grab for receiving; Receive and correspond to by the second additional character signal of second microphone at the second digitized signal stream of the voice signal of very first time frame-grab; Represent based on described first additional character calculated signals the 3rd T/F; Represent based on described second additional character calculated signals the 4th T/F; To represent with the 4th T/F upgrade the amplitude calibration factor with representing based on the 3rd T/F.

In certain embodiments, this device, method and/or non-transitory computer-readable medium can comprise module, step or executable instruction, for identifying amplitude that the 3rd T/F of very first time frame the represents frequency lower than noise level, and when upgrading amplitude correction factor based on the 3rd time-frequency representation, abandon the 3rd time-frequency representation of identified frequency and very first time frame.

In certain embodiments, device, method, and/or described non-transitory computer-readable medium can comprise module, step or executable instruction, for identifying that the 3rd T/F of described very first time frame represents and the frequency that undesirable acoustical signal is associated; With when upgrading amplitude correction factor based on the 3rd time-frequency representation, abandon the 3rd time-frequency representation of identified frequency and very first time frame.

In certain embodiments, this device, method and/or described non-transitory computer-readable medium can comprise module, step or executable instruction, for be enough different from when the ratio that the 4th time-frequency representation and the 3rd T/F represent based on the very first time-frequency representation and the second T/F represent the amplitude correction factor calculated time, when determining that described 3rd time-frequency representation is associated with underproof acoustical signal.

In certain embodiments, T/F represents and comprises the one or more of short time discrete Fourier transform (STFT) or wavelet transformation.

In certain embodiments, this device can comprise the interface being configured to reception first digitized signal stream and the second digitized signal stream, wherein, described first digitized signal stream and the second digitized signal stream correspond to the voice signal of being caught respectively by the first microphone and second microphone.This device can also comprise the processor with described interface communication, is configured to run the module stored in memory.This module can be configured to determine described first digitized signal stream the very first time-the second T/F of frequency representation and the second digitized signal stream represents, wherein, the very first time-frenquency representation indicates the first digitized signal stream relative to the amplitude of multiple frequencies of very first time frame, and described second T/F representation represents the amplitude of the second digitized signal stream relative to described multiple frequency and very first time frame.It is positioned opposite that this module also can be configured to based on described first microphone and second microphone, and more than first frequency and very first time frame the very first time-frequency representation and the second T/F represent and calculate the first parameter of the arrival direction of instruction voice signal.This module also can be configured to based on more than first frequency and very first time frame the first parameter, the very first time-frequency representation and the second time-frequency representation, determine the first relative phase-angle error for very first time frame and more than first frequency between described first microphone and second microphone.

In certain embodiments, the method can comprise: receive the first digitized signal stream and the second digitized signal stream by the data processing module being coupled to the first microphone and second microphone, wherein, described first digitized signal stream and the second digitized signal stream corresponding voice signal of being caught by the first microphone and second microphone respectively.The method can also comprise: data processing module determine described first digitized signal stream the very first time-the second T/F of frequency representation and the second digitized signal stream represents, wherein, the very first time-frequency representation indicates the first digitized signal stream for the phase place of multiple frequency and very first time frame, and wherein said second T/F represent instruction second digitized signal stream for described multiple frequency with the phase place of very first time frame.The method can also comprise: the calibration module communicated with data processing module based on described first microphone and described second microphone positioned opposite, more than first frequency and very first time frame the very first time-frequency representation and the second T/F represent and calculate the first parameter, the arrival direction of its instruction acoustical signal.The method can also comprise: at calibration module, based on more than first frequency and very first time frame the first parameter, the very first time-frequency representation and the second time-frequency representation, and determine the first microphone and second microphone the first relative phase-angle error for very first time frame and more than first frequency.

In certain embodiments, non-transitory computer-readable medium can comprise executable instruction, can operate to impel data processing equipment to pass through to be coupled to interface first digitized signal stream and the second digitized signal stream of the first microphone and second microphone, wherein, the first digitized signal stream and the second digitized signal stream correspond to the voice signal caught respectively by described first microphone and described second microphone.Described computer-readable medium can also comprise executable instruction, can operate to make data processing equipment determine described first digitized signal stream the very first time-the second T/F of frequency representation and the second digitized signal stream represents, wherein, the very first time-frequency location represents the phase place of the first digitized signal stream for multiple frequency and very first time frame, and wherein said second T/F represents the phase place of instruction second digitized signal stream for multiple frequency and very first time frame.Described computer-readable medium can also comprise executable instruction, can operate to impel data processing equipment based on described first microphone and second microphone positioned opposite, represent more than first frequency with at a time frequency representation of very first time frame and the second T/F and calculate the first parameter of the arrival direction of instruction voice signal.Described computer-readable medium can also comprise executable instruction, can operate to make data processing equipment based on the first parameter, more than first frequency and very first time frame the very first time-frequency representation and the second temporal frequency represent and determine the first relative phase-angle error for very first time frame and more than first frequency between the first microphone and second microphone.

In certain embodiments, this installation method and/or described non-transitory computer-readable medium can comprise module, step or executable instruction, for determine sampling frequency multiple described in first and very first time frame described the very first time-frequency representation and the second T/F represent between first-phase potential difference; And determine the first parameter based on described first-phase potential difference.

In certain embodiments, in this device, the method and/or described non-transitory computer-readable medium can comprise module, step or executable instruction, for determining the first parameter according to linear system, described linear system relate at least partly arrival direction and the very first time-phase difference between frequency representation and the second T/F representation.

In certain embodiments, this device, method and/or non-transitory computer-readable medium can comprise module, step or executable instruction, for receiving the first additional character signal of the first digitized signal stream corresponding to the voice signal of being caught at the second time frame by the first microphone; Receive the second additional character signal of the second digitized signal stream corresponding to the voice signal of being caught at the second time frame by second microphone; Based on described first additional character signal, the 3rd T/F calculating the second time frame represents; Based on described second additional character signal, the 4th T/F calculating the second time frame represents; Based on the first relative phase-angle error of positioned opposite, the very first time frame for the 3rd frequency representation of the second time frame and the 4th frequency representation, the first microphone and second microphone, determine to represent second parameter of described acoustical signal for the arrival direction of the second time frame; With based on the 3rd frequency representation and the 4th frequency representation in the second time frame and the second parameter, determine that the first microphone and second microphone are for the second relative phase-angle error between the second time frame and more than first frequency.

In certain embodiments, this device, method and/or described non-transitory computer-readable medium can comprise module, step or executable instruction, for determining the second relative phase-angle error based on described first relative phase-angle error, with relative to described first relative phase-angle error smoothly described second relative phase-angle error.

In certain embodiments, this device, method and/or described non-transitory computer-readable medium can comprise module, step or executable instruction, for determine when the first parameter and the second parameter closer to each other time the second relative phase-angle error time, the discretization of the arrival direction of described first parameter instruction very first time frame, and the second parameter indicates the discretization of the arrival direction of the second time frame.

In certain embodiments, this device, method and/or described non-transitory computer-readable medium can comprise module, step or executable instruction, and for providing mask, the amplitude of this mask identification the 3rd time-frequency representation is less than the frequency of noise level.

In certain embodiments, this device, method and/or described non-transitory computer-readable medium can comprise module, step or executable instruction, for using this mask to abandon the 3rd time-frequency representation to identified frequency in estimation second relative phase-angle error.

In certain embodiments, this device, method and/or non-transitory computer-readable medium can comprise module, step or an executable instruction, for providing identification the 3rd frequency mask that time-frequency representation is associated with undesirable acoustical signal.

In certain embodiments, this device, method and/or described non-transitory computer-readable medium can comprise module, step or executable instruction, for using this mask to abandon the 3rd time-frequency representation to identified frequency in estimation second relative phase-angle error.

In certain embodiments, this device, method and/or non-transitory computer-readable medium can comprise module, step or executable instruction, at least two the first relative phase-angle errors be associated that are level and smooth and described multiple frequency.

In certain embodiments, this device, method and/or described non-transitory computer-readable medium can comprise module, step or executable instruction, for receiving the first additional character signal of the first digitized signal stream corresponding to the voice signal of being caught at the second time frame by the first microphone; The 3rd T/F calculating the second time frame based on described first additional signal digitlization represents; The first relative phase-angle error is removed, to calibrate the first microphone for frequency multiple described in first relative to described second microphone with from the 3rd time-frequency representation for more than first frequency and the second time frame.

Disclosed calibration steps (it comprises device described herein, system and method) can provide following one or more advantage.Disclosed collimation technique can estimate the alignment profiles line that microphone is online, such as, when microphone is disposed in practical operation.Therefore, disclosed collimation technique does not need to dispose in test environment, and this may be consuming time and expensive.Disclosed collimation technique also can be deployed in offline session, such as, during independent calibration session.Disclosed collimation technique can estimate the phase error variations compensated for the amplitude calibration Summing Factor relative phase-angle error of the change of compensation magnitude sensitivity.In addition, disclosed collimation technique can even use when multi-acoustical exists.As described below, disclosed collimation technique system can eliminate any deviation introduced by multi-acoustical, and does not initiatively abandon signal from multi-acoustical.

Therefore the feature of disclosed theme is summarized quite widely, to understand following detailed description better, and in order to understand the contribution to prior art better.Certainly, this will be described below and will form the subject content of appended claims.

Accompanying drawing explanation

In conjunction with accompanying drawing below, with reference to the following detailed description of disclosed theme, better can understand various objects, the feature and advantage of disclosed theme, wherein identical reference number represents identical element.

Fig. 1 illustrates according to the relation between the input audio signal of some embodiments and the signal of telecommunication of detection.

Fig. 2 illustrates according to some embodiments, wherein can use the installation of calibrating installation or system.

Fig. 3 illustrates according to some embodiments, and how the signal detected is further processed to calibrate microphone.

Fig. 4 illustrates the Data Preparation Process of the data preparation module according to some embodiments.

Fig. 5 shows the amplitude calibration process of the amplitude calibration module of the amplitude sensitivity according to some embodiments calibration microphone.

Fig. 6 A-6B illustrates the amplitude proportional histogram h according to some embodiments _i(ω, r).

Fig. 7 shows arrival direction θ and the phase error phi of microphone _i(ω) phase difference between observation signal how is caused.

Fig. 8 A-8B illustrates according to some embodiments for solving the process of linear equation system.

Fig. 9 A-9C illustrates the progress of amplitude according to some embodiments and phase calibration process.

Figure 10 A-10D shows the advantage according to the correcting mechanism calibration microphone disclosed in the use of some embodiments.

Figure 11 shows according to some embodiments, uses auto-adaptive filtering technique to estimate the method for alignment profiles.

Figure 12 is the block diagram of the computing equipment according to some embodiments.

Figure 13 A-13B illustrates according to some embodiments for one group of microphone in conjunction with disclosed calibration process.

Figure 14 illustrates according to some embodiments, is determined the process of amplitude correction factor by the relation between the time-frequency representation of estimating the input acoustical signal received at multiple time frame.

Figure 15 illustrates according to some embodiments, relates to the exemplary scatter diagram that the time-frequency representation corresponding to same time frame is sampled.

Embodiment

In the following description, the method for regarding system and disclosed theme and many details of the manipulable environment of wherein such system and method are set forth, in order to provide the theme disclosed in thorough understanding.But, it will be apparent to those skilled in the art that disclosed theme can not have to put into practice when these details, and some feature as known in the art is not described in detail, in order to avoid the theme disclosed in complicated description.In addition, should be appreciated that, the embodiment provided below is exemplary, and can be expected that to also have other system and method within the scope of disclosed theme.

Microphone comprises and is configured to receive voice signal s (t), and converts thereof into the transducer of signal of telecommunication m (t), and wherein t represents time variable.Ideally, microphone has smooth frequency domain transfer function:

H(ω)＝A

Wherein A is conversion gain coefficient.Therefore, for all frequencies paid close attention to, desirable microphones voice signal, and convert thereof into the signal of telecommunication and without any delay.

Unfortunately, typical microphone shows certain non-ideal characteristic.Such as, microphone can be deferred to voice signal m (t) of conversion relative to the interpolation of described input audio signal s (t).Fig. 1 illustrates according to the relation of some embodiments between input acoustical signal s (t) and signal of telecommunication m (t) 104 detected.Because the non-ideal characteristic of microphone, signal of telecommunication m (t) detected postpones relative to described input audio signal s (t) 102 to postpone Δ t.

In addition, characteristic (such as, the conversion gain factors A and/or to postpone Δ t can be frequency dependence of microphone.Such as, when microphone with 0.8 the conversion gain factor to decay the acoustical signal of 10 kilo hertzs, identical microphone to be decayed the acoustical signal of 15 kilo hertzs by the conversion gain factor of 0.7.Equally, although microphone postpones 10 kilo hertzs of acoustic signals with 0.1 millisecond, same microphone can postpone 15 kilo hertzs of acoustic signals with 0.1 millisecond.Therefore, the transfer function with the imperfect microphone that frequency dependence conversion gain factor and frequency dependence postpone can modeling as follows:

H(ω)＝A(ω)exp(iφ(ω))

Wherein A (ω) indicates the frequency dependent conversion gain factor; instruction corresponds to the frequency-dependant phase error of time delay Δ t; And

If all microphones have identical nonideal characteristic, the non-ideal characteristic of microphone is not also a problem, because the great majority application supposition microphone of multiple microphone is imperfect, but nonideal mode is identical.But because uncontrolled change in the fabrication process, different microphones has different characteristics, and it can cause the mistake of the application of the same characteristic features depending on microphone.

Manufacture change to solve, various collimation technique has been developed the phase error phi (ω) estimating conversion gain factors A (ω) and microphone.The estimated phase error estimated by conversion gain Summing Factor can be used for passing through compensating filter c (t) of the following transfer function had in frequency domain and removes the impact of the transfer function of microphone from detected signal m (t):

$C\left(\mathrm{\ω}\right)=\frac{1}{A\left(\mathrm{\ω}\right)}\exp (-i\mathrm{\φ}(\mathrm{\ω}\left)\right).$

Like this, be constant in the overall transfer function of microphone and compensating filter for all frequencies, thus approach desirable microphone:

$H\left(\mathrm{\ω}\right)C\left(\mathrm{\ω}\right)=A\left(\mathrm{\ω}\right)\exp \left(i\mathrm{\φ}\right(\mathrm{\ω}\left)\right)\×\frac{1}{A\left(\mathrm{\ω}\right)}\exp (-i\mathrm{\φ}(\mathrm{\ω}\left)\right)=1.$

One class collimation technique is called as calibrated off-line technology.Calibrated off-line technical testing uses the calibration sound source of given frequency in anechoic chamber, test microphone and measures the response of microphone to calibration sound source.This step can iteration be the different sound sources with different frequency, to determine alignment profiles C (ω), for the frequency of each concern.The advantage of calibrated off-line technology is: it can provide the accurate calibration profile of microphone.But off-line calibration technology is consuming time and non-economy, because each microphone must be tested for the frequency of each concern.In addition, off-line calibration technology can not explain other similar modification of the aging and microphone characteristics of the microphone due to time or use, because calibration only performed once usually before initial use.

Another kind of collimation technique is called as on-line calibration technology.When microphone is deployed in true environment, on-line calibration technology can provide the alignment profiles of the microphone using the signal detected.In order to reduce the dimension of problem, on-line calibration technology estimates Relative Transformation gain coefficient (instead of conversion gain factors A (ω)) or relative phase-angle error (replacing phase error phi (ω)) usually.Even if having the minimizing of dimension, most of on-line calibration technology can only estimate the relative conversion gain factor, instead of relative phase-angle error.In addition, can estimate that a small amount of on-line calibration technology of relative conversion gain Summing Factor relative phase-angle error is carried out about sound source hypothesis strict especially.Such as, the U.S. Patent number 8243952 being " MicrophoneArrayCalibrationMethodandApparatus " by Thormundsson, title illustrate for estimate two microphones (or more) between the method for relative phase-angle error, by only having when sound source upgrades relative phase-angle error when the front of two microphones completely.If not impossible, because be difficult to estimate that estimated relative phase-angle error can be inaccurate when sound source is completely before two microphones.

Disclosed device, system and method are provided for the collimation technique calibrating microphone group.As long as because microphone has substantially the same feature, the great majority application of multi-microphone system can hold imperfect microphone, and disclosed collimation technique is configured to relative datum Microphone calibration microphone.Disclosed technology is specially adapted to calibrate omnidirectional and fully close one group of microphone each other.The alignment profiles in frequency domain can be expressed relative to the calibration result of the microphone of reference microphone:

Wherein represent the conversion gain factors A corresponding to i-th microphone _i(ω) and correspond to the conversion gain factors A of reference microphone _r(ω) ratio between; And represent the relative phase-angle error between two microphones, λ _i(ω) the amplitude calibration factor of i-th microphone is also referred to as.

Disclosed correcting mechanism can comprise or use two modules: amplitude calibration module and phase alignment module.Amplitude calibration module is configured to the amplitude correction factor λ of the microphone of the reference microphone determined relative to each frequency _i(ω).When microphone is fully close to time each other, will be enough identical by the voice signal of microphones.Therefore, can owing to the amplitude correction factor of microphone in any difference of the signal detected by microphone.

Therefore, amplitude calibration module is configured to determine that the T/F of the signal detected by microphone represents that (TFR) and their TFR of calculating are in the ratio paying close attention to frequency, and this is that the amplitude calibration factor lambda between frequency paid close attention to by microphone in theory _i(ω).But due to noise and other non-ideal characteristics of microphone, a sample of TFR ratio may be not accurate enough, because amplitude calibration factor lambda _i(ω) estimation.Therefore, in order on average fall noise and other non-ideal characteristics, amplitude calibration module is configured to be collected in the many TFR samples paying close attention to frequency, and estimates the amplitude calibration factor from TFR sample.

In certain embodiments, amplitude calibration module is configured to create the histogram of described TFR ratio at the sample of interested frequency, and estimates the amplitude calibration factor from histogram.When the signal additional sample that microphones is detected by microphone, amplitude calibration module can use other sample to calculate the other sample of TFR ratio, comprise the available sample of other sample to TFR ratio of TFR ratio, and reappraise the amplitude calibration factor of the sample of the TFR ratio in the group after based on renewal.Because amplitude correction factor can be reappraised as Received signal strength additional sample, amplitude calibration module can follow the tracks of the time-varying characteristics of microphone, because aging and/or prolongation use.

In certain embodiments, amplitude calibration block configuration becomes by determining that the relation corresponded between the TFR sample of same time frame estimates the amplitude calibration factor.Such as, amplitude calibration module can suppose that the relation between TFR sample is linear.Therefore, amplitude calibration module can estimate the amplitude calibration factor by identifying the line of the relation represented between TFR sample.

In certain embodiments, phase alignment module is configured to the relative phase error determining i-th microphone relative to the benchmark microphone of each frequency observation phase difference between the signal detected by two microphones can depend on the arrival direction of (1) input audio signal and the relative phase-angle error of (2) microphone therefore, phase alignment module is configured to the phase difference estimation arrival direction observed between the signal that detected by two microphones and relative phase-angle error.In some cases, phase alignment block configuration is repeatedly estimate arrival direction and relative phase-angle error one by one.When phase alignment module receives the other sample of observed phase difference in time, phase alignment module can upgrade the estimated value of arrival direction and relative phase-angle error further.Because when receive detect the additional sample of voice signal time again can estimate relative phase-angle error, phase alignment module also can follow the tracks of the time-varying characteristics of microphone.Because aging and/or prolongation use.

Even when multi-acoustical exists, disclosed collimation technique can be used.As described below, disclosed collimation technique system can eliminate any deviation that superposition source and near field sources introduce, and reduces the quantity of the data sample abandoned.

In certain embodiments, disclosed collimation technique can as off-line calibration mechanism.Such as, user can use integrated microphone to test the microphone of quiet environment, such as cell phone in the electronic device, and uses amplitude calibration module and phase alignment module to estimate the alignment profiles of microphone.

In certain embodiments, the alignment profiles of microphone can be represented as centrifugal pump.In this discrete representation of alignment profiles, Ω can represent frequency range in a frequency domain.In certain embodiments, benchmark microphone can be one of microphone calibrated.In some cases, one group of microphone that disclosed collimation technique can be used for from being calibrated is selected from reference microphone.In certain embodiments, alignment profiles can be expressed as microphone impulse response in the time domain.

Fig. 2 illustrates the scene that can use disclosed correcting mechanism according to some embodiments.Fig. 2 comprises the sound source 202 producing acoustical signal s (t).Acoustical signal s (t) can be propagated to (i+1) microphone 204A-204E by propagation medium, and wherein i can be any value being more than or equal to 1.

If (minimum range between being expressed as l) is greater than in fact the ultimate range d between microphone, and so acoustical signal s (t) can be approximated to be greatly one-way planar ripple 206 substantially for microphone and described sound source 202.Such as, the distance between microphone can be limited in 2-3mm, and it significantly can be less than the wavelength of input audio signal s (t) or the minimum range between microphone and described sound source.As another example, distance between microphone can be at centimetres, this is still significantly less than minimum range in many application scenarioss between microphone and the described sound source microphone of Set Top Box (in the living room of such as, the sound instruction of recipient).

Microphone 204 can receive described acoustical signal s (t), and converts thereof into the signal of telecommunication.For illustrative purposes, the signal of telecommunication detected by reference microphone is called as m _r(t); The signal of telecommunication detected by other microphones is called as m ₁(t) ... m _i(t).Microphone 204 can provide detected signal m ₁(t) ... m _i(t), m _r(t) to back end computing device (not shown), with based on the signal m detected ₁(t) ... m _i(t), m _rt (), this computing equipment can determine the alignment profiles of i-th microphone relative to described benchmark microphone.

Although Fig. 2 only comprises a sound source, disclosed correcting mechanism can be combined the sound source of any number simultaneously launching sound.Disclosed technology also can be combined any layout of microphone.Such as, in certain embodiments, microphone can be arranged to array (such as, along straight line); In other embodiments, microphone can be arranged to random shape.

Fig. 3 shows the signal how to detect and is processed further according to some embodiments by a back end computing device.Fig. 3 comprises sound source 202, one group of microphone 204, analog to digital converter (ADC) 302, data preparation module 304, the calibration module 306 comprising amplitude calibration module 308 and phase alignment module 310 and application module 312.One group of microphone 204 can provide detected signal m ₁(t) ... m _i(t), m _rt () is to ADC302, and ADC302 can provide digitized signal to data preparation module 304.This digitized signal is also referred to as m ₁[n] ... m _i[n], m _r[n], wherein n can refer to the frequency range (such as, the wherein time range of ADC302 sample detecting signal, or time frame) in time domain.Digitized signal also can be called as digitized signal stream, because digitized signal can comprise the sample of signal corresponding to different time frame.

Data preparation module 304 can calculate digitized signal M ₁[n, Ω] ... M _i[n, Ω], M _rthe T/F of [n, Ω] represents (TFR).The TFR of digitized signal can be associated with multiple discrete frequency frequency range and multiple discrete time frequency range.Such as, M _i[n, the Ω] of [n, Ω] refers to the T/F frequency range (or index) of discrete time-frequency domain.In certain embodiments, the size of described multiple discrete frequency frequency range can be identical.In other embodiments, the size of described multiple discrete frequency frequency range can be different from each other, such as, with layering time-frequency representation.Equally, in certain embodiments, described multiple discrete time frequency range can be identical size; In other embodiments, the size of described multiple discrete time frequency range can be different from each other.Can in predetermined frequency range and the scope of time that is associated with each T/F frequency range.TFR corresponding to the digitized signal of time frame is called as a sample or a data sample.Time-frequency representation can comprise short time discrete Fourier transform (STFT), wavelet transformation, linear-elastic buckling, fractional fourier transform, New World conversion, constant Q transform and Gabor transformation.In some cases, time-frequency representation can be generalized to any linear transformation of the window part putting on measured signal further.

Data preparation module 304 can also use the alignment profiles of the previous estimation of i-th microphone to compensate i-th amplitude calibration Summing Factor relative phase-angle error between microphone and reference microphone, thus provides digitized signal calibration TFR.

Data preparation module 304 can provide digitlization switching signal M subsequently ₁[n, Ω] ... M _i[n, Ω], M _r[n, Ω] is to calibration module 306 and digitlization switching signal calibration TFR to application module 312.

Calibration module 306 can use amplitude calibration module 308 and phase alignment module 310 to use the digitlization switching signal M received by calibration module 306 ₁[n, Ω] ... M _i[n, Ω], M _rthe additional TFR sample of [n, Ω] and reappraise the alignment profiles of microphone.Calibration module 306 can provide subsequently and reappraise alignment profiles to data preparation module 304, and the TFR subsequently of the signal that digitlization is changed can use and recalibrate estimation curve calibration.On the other hand, in various applications, application module 312 can process the calibration TFR of the digitized signal received from data preparation module 304.In certain embodiments, calibration module 306 can provide the alignment profiles of microphone to application module 312, alignment profiles process can be used to import digitized signal into make application module 312.

Fig. 4 illustrates the Data Preparation Process of the data preparation module according to some embodiments.In step 402, data preparation module 304 can receive i+1 digitized signal m from ADC304 ₁[n] ... m _i[n], m _r[n], and calculate digitlization switching signal M ₁[n, Ω] ... M _i[n, Ω], M _rthe TFR of [n, Ω].Such as, data preparation module 304 can calculate i+1 digitized signal m ₁[n] ... m _i[n], m _rthe Discrete Short Time Fourier transform of [n].The T/F resolution of D-STFT can be depending on predetermined time/parameter of frequency resolution.This predetermined analytic parameter can be depending on available amount of ram and maintains calibrating section and/or the right resolution desired by application module 312 of signal.In certain embodiments, data preparation module 304 can a sequence reception i+1 digitized signal m ₁[n] ... m _i[n], m _r[n].In this case, data preparation module 304 can order computation digitlization conversion signal M ₁[n, Ω] ... M _i[n, Ω], M _rthe TFR of [n, Ω], is similar to bank of filters.Such as, when data preparation module 304 receives the special time frame that new digitized signal is particular microphone, data preparation module 304 can calculate special time frame TFR's and row are added to corresponding to the existing TFR of time frame before particular microphone.

In step 404, data preparation module 304 optionally identifies that the amplitude had is the data sample lower than noise level.Such as, data preparation module 304 can receive noise variance parameter, and instruction microphone has σ ²noise variance.If a order of magnitude [n=n of the TFR of target microphone (such as, microphone is calibrated) ₀, Ω=Ω ₀], M _i[n=n ₀, Ω=Ω ₀] being less than σ, then data preparation module 304 can identify the specific sample M of TFR _i[n=n ₀, Ω=Ω ₀] too noisy.If the size M of the TFR of reference microphone _r[n=n ₀, Ω=Ω ₀] being less than σ, then data preparation module 304 can identify all data sample M ₁[n=n ₀, Ω=Ω ₀] ..., M _i[n=n ₀, Ω=Ω ₀], M _r[n=n ₀, Ω=Ω ₀] too noisy, because M _r[n=n ₀, Ω=Ω ₀] all microphones of calibration estimation can be affected.In certain embodiments, data preparation module 304 can represent the noise data sample identified using mask.Such as, mask can have the same dimension of the TFR of digitlization switching signal, and whether the data sample that instruction corresponds to the frequency range of mask has the amplitude being less than noise level.

In a step 406, data preparation module 304 optionally identifies that corresponding to voice signal does not meet the data sample supposed in plane wave, single source.Undesirable acoustical signal can comprise the acoustical signal corresponding to and receive from near-field sound source, in conjunction with the signal of the acoustical signal of multi-acoustical, or because reverberation source is corresponding to the acoustical signal of reverberation.Such as, near-field sound source is physically located at the sound source near microphone.When sound source is near microphone, input audio signal is no longer plane wave.Therefore, the acoustical signal received is that the hypothesis of plane wave can not keep near-field sound source.

In order to determine sample M _i[n=n ₀, Ω=Ω ₀] be associated with underproof acoustical signal, described data preparation module 304 can for the concern ratio of frequency computation part i-th between microphone and the signal amplitude of reference microphone:

$r_i\[n_0,{\mathrm{\Ω}}_0\]=\frac{\left|\right|M_R\[n_0,{\mathrm{\Ω}}_0\]\left|\right|}{\left|\right|M_i\[n_0,{\mathrm{\Ω}}_0\]\left|\right|},$

And if this ratio r _i[n ₀, Ω ₀] be enough different from amplitude calibration factor lambda _ithe current estimation of [Ω], then data preparation module 304 can indicate particular data sample M _i[n ₀, Ω ₀] be associated with undesirable acoustical signal.

In certain embodiments, when particular data sample meets following relationship, data preparation module 304 can indicate particular data sample to be associated with any one near-field sound source or multi-acoustical:

||λ _i[Ω ₀]-r _i[n ₀，Ω ₀]||＞δ ₀

Wherein δ _dit is predetermined threshold.In other embodiments, when particular data sample meets following relationship, data preparation module 304 can indicate particular data sample to be associated with any one near-field sound source or multi-acoustical:

$max(\frac{\left|\right|{\mathrm{\λ}}_i\[{\mathrm{\Ω}}_0\]\left|\right|}{\left|\right|r_i\[n_0,{\mathrm{\Ω}}_0\]\left|\right|},\frac{\left|\right|r_i\[n_0,{\mathrm{\Ω}}_0\]|}{\left|\right|{\mathrm{\λ}}_i\[{\mathrm{\Ω}}_0\]\left|\right|})>{\mathrm{\δ}}_R,$

Wherein δ _dit is predetermined threshold.

In certain embodiments, data preparation module 304 can use the data sample that mask identification is associated with undesirable acoustical signal.Mask can have identical dimension with the TFR of the signal that digitlization is changed, and whether the data sample that instruction corresponds to the frequency range of mask is associated with any one near-field sound source or multi-acoustical.Data preparation module 304 can provide mask to other modules, such as calibration module 306 or application module 312, mask can be used to improve their operational quality to make other modules.Such as, application module 312 can use mask to improve the performance of blind source separating.In certain embodiments, before providing data sample to calibration module 306 or application module 312, data preparation module 304 can abandon the data sample be associated with any one near-field sound source or multi-acoustical.

In certain embodiments, for can be suitable for the environment be deployed according to microphone wherein from the predetermined threshold of undesirable acoustical signal processing data sample.Such as, whether different predetermined thresholds can be deployed between outdoor, indoor, meeting, meeting room, living room, big room, cubicle, resting room or in automobile according to microphone uses.In some cases, predetermined threshold can use supervised learning technology (such as, returning) and learn.

In a step 408, data preparation module 304 optionally can estimate the parameter indicating described input audio signal s (t) arrival direction (DOA).The parameter of expression DOA can be DOA itself, but also can be any parameter of relevant DOA or approximate DOA.Represent that the parameter of DOA can be described as DOA designator, or in this application simply as DOA.In some cases, estimated parameter can by application module 312 for application.Estimated parameter also can by phase alignment module 310 for estimating the alignment profiles of relative phase-angle error.In certain embodiments, DOA indicating device can by phase alignment module 310, instead of data preparation module 304 is estimated.

In certain embodiments, DOA indicating device can use Multiple Signal Classification (MUSIC) method to estimate.In other embodiments, the spendable ESPRIT method of DOA indicating device is estimated.In certain embodiments, DOA indicating device can use beam-forming method to estimate.

In certain embodiments, the DOA indicating device of input audio signal is estimated by solving linear equation system:

$\left[\begin{array}{c}{\mathrm{\η}}_1^T\[\mathrm{\Ω},\mathrm{\θ}\]\\ ...\\ {\mathrm{\η}}_i^T\[\mathrm{\Ω},\mathrm{\θ}\]\end{array}\right]=2\mathrm{\π}\frac{{\mathrm{\Ωf}}_s}{2P}v\left[\begin{array}{c}-r_1-\\ ...\\ -r_i-\end{array}\right]\left[\begin{array}{c}cos\mathrm{\θ}\\ sin\mathrm{\θ}\end{array}\right]$

Wherein, due to DOA indicating device θ, the relative phase delay between i-th microphone and reference microphone (such as, at time frame T), f _sbe the sample frequency of ADC302, Ω is frequency range in a frequency domain, and P represents the quantity (such as, resolution) of the frequency window of T/F conversion (such as, STFT), and v is the speed of described acoustical signal, r _ibe represent the two-dimensional vector of i-th microphone relative to the position of reference microphone, and θ is the DOA indicating device of voice signal.The said system of linear equation relates to the delay between signal and the DOA indicating device of acoustical signal detected by microphone.Relative phase delay can be depending on the relative position of microphone, it can be caught by bivector ri.Based on frequency and the speed of described input audio signal, all the other systems of linear equation can postpone for phase delay change-over time.In certain embodiments, f _s, Ω and P can be merged into single item, represents the discrete frequency of the input acoustical signal measured by microphone.

In certain embodiments, relative phase delay can measured or calculating.Such as, phase delay the TFR value that can associate with reference microphone by comparing the i-th microphone calculates.Particularly, phase delay can calculate as follows:

${\mathrm{\η}}_i^T\[\mathrm{\Ω},\mathrm{\θ}\]=\arg (M_i\[n=T,\mathrm{\Ω}\])-\arg (M_R\[n=T,\mathrm{\Ω}\])$

Wherein arg provides the angle of complex variable.

This linear system can adopt linear system solution device to solve relative to θ.Because this equation be complete system (such as, equation system comprises the more constraint than unknown number), as z > 1, linear system can utilize least square method to solve: find the θ reducing overall minimum side's error.In certain embodiments, linear system can use matrix $2\mathrm{\π}\frac{{\mathrm{\Ωf}}_s}{2P}v\left[\begin{array}{c}-r_1-\\ ...\\ -r_i-\end{array}\right]$ A mole Roger Penrose pseudoinverse solve.Therefore, solve linear system and can relate to below calculating:

${\left(2\mathrm{\π}\frac{{\mathrm{\Ωf}}_s}{2P}v\left[\begin{array}{c}-r_1-\\ ...\\ -r_i-\end{array}\right]\right)}^{\⊥}\left[\begin{array}{c}{\mathrm{\η}}_1\[\mathrm{\Ω},\mathrm{\θ}\]\\ ...\\ {\mathrm{\η}}_i\[\mathrm{\Ω},\mathrm{\θ}\]\end{array}\right]=\left[\begin{array}{c}cos\mathrm{\θ}\\ sin\mathrm{\θ}\end{array}\right],$

Wherein, ⊥ represents a mole Roger Penrose pseudoinverse.

In certain embodiments, data preparation module 304 can use the calibration curve previously calculated to compensate amplitude correction factor and the relative phase-angle error of microphone.Data preparation module 304 can by being multiplied by the TFR of the digitlization switching signal of the i-th microphone and corresponding alignment profiles and compensation magnitude/phase error:

$\begin{array}{c}{\hat{M}}_1\[n,\mathrm{\Ω}\]=F_1\[\mathrm{\Ω}\]M_1\[n,\mathrm{\Ω}\]\\ ...\\ {\hat{M}}_i\[n,\mathrm{\Ω}\]=F_i\[\mathrm{\Ω}\]M_i\[n,\mathrm{\Ω}\]\\ {\hat{M}}_R\[n,\mathrm{\Ω}\]=M_R\[n,\mathrm{\Ω}\]\end{array}$

Wherein F _i[Ω] refers to i-th estimation of the alignment profiles of i-th microphone.

Then, data preparation module 304 can provide the TFRM of digitlization switching signal ₁[n, Ω] ... M _i[n, Ω], M _rthe calibration TFR of the signal of [n, Ω], digitlization conversion identify noise data sample the first mask and/or identify and the second mask of data sample that data sample that near-field sound source or multi-acoustical are associated is associated to calibration module 306 and/or application module 312.

The signal M that calibration module 306 can use digitlization to change ₁[n, Ω] ... M _i[n, Ω], M _r[n, Ω] estimates the alignment profiles of microphone in discrete frequency domain:

Wherein represent i-th size calibration factor between microphone and reference microphone, and represent i-th relative phase-angle error between microphone and reference microphone.

Fig. 5 shows the amplitude sensitivity of how to calibrate microphone according to some embodiment amplitude calibration modules.Amplitude calibration module 308 can suppose that microphone is close to each other.The possibility that amplitude calibration module 308 can also suppose to occupy the different sound sources of same time-frequency bins in time-frequency representation is very little.This hypothesis is often met, because different sound sources has different frequency characteristics usually.

Under these assumptions, if i-th microphone and reference microphone have the sensitivity of the identical order of magnitude, described input audio signal M _i[n, Ω] and M _rthe amplitude of the TFR of [n, Ω] is identical.Therefore, the signal M detected _i[n, Ω] and M _rany amplitude difference between the TFR of [n, Ω] can owing to the difference of the amplitude sensitivity at this special time-frequency bins.Amplitude calibration module 308 can use this characteristic to estimate the amplitude calibration factor.

In step 502, amplitude calibration module 308 can calculate TFRM _i[n, Ω] and M _rthe ratio of the amplitude of [n, Ω]:

$r_i\[n,\mathrm{\Ω}\]=\frac{\left|\right|M_R\[n,\mathrm{\Ω}\]\left|\right|}{\left|\right|M_i\[n,\mathrm{\Ω}\]\left|\right|}.$

In certain embodiments, amplitude calibration module 308 can use the mask provided by data preparation module 304, to remove noisy TFR sample, or the TFR sample be associated with arbitrary near-field sound source or multi-acoustical.

In step 504, amplitude calibration module 308 can in time n for frequency bins Ω ₀collect two or more ratio, to determine the summary info of ratio.It is useful information for the amplitude correction factor determined that the summary info of ratio can indicate.

In certain embodiments, summary info can comprise i-th microphone for characteristic frequency frequency range Ω ₀the block diagram of ratio:

$h_i^T\[{\mathrm{\Ω}}_0,r\]=hist(r_i\[n,{\mathrm{\Ω}}_0\]),n=1...T$

Wherein, T is proportional sample r _i[n, Ω ₀] available up-to-date time frame, and r represents ratio amplitude.Histogram is the expression at row frequency-distributed interval (frequency range), and wherein, described frequency representation falls into multiple ratios at interval.

Fig. 6 A-6B shows the histogram according to some embodiments fig. 6 A illustrates histogram as image, wherein line display frequency axis, and amplitude axis is shown in list.Histogram multiple samples in special frequency channel [Ω, r].Fig. 6 B illustrates the cross section of image in Ω=250 of Fig. 6 A:

$h_i^T\[\mathrm{\Ω}=250,r\].$

In step 506, amplitude calibration module 308 can use summary info to estimate the amplitude calibration factor in certain embodiments, amplitude calibration module 308 is by calculating the M of TFR _i[n, Ω] and M _rthe intermediate value of the ratio of [n, Ω] and estimate amplitude correction factor:

$r_i\[n,\mathrm{\Ω}\]=\frac{\left|\right|M_R\[n,\mathrm{\Ω}\]\left|\right|}{\left|\right|M_i\[n,\mathrm{\Ω}\]\left|\right|}.$

In other embodiments, when summary info comprises the histogrammic of ratio, amplitude calibration module 308 can operate estimation histogram f () estimate amplitude calibration factor lambda _i[Ω]:

${\widetilde{\mathrm{\λ}}}_{i,T}\[\mathrm{\Ω}\]=f(h_i^T\[\mathrm{\Ω},r\])$

Wherein represent amplitude correction factor λ _ithe estimation of [Ω], and wherein subscript T represents based on until the amplitude calibration factor of sample that receives of time frame T.

In certain embodiments, estimator f () can be configured to be identified in histogram h _ithere is in [Ω, r] ratio of the sampling of maximum quantity:

${\widetilde{\mathrm{\λ}}}_{i,T}\[\mathrm{\Ω}\]=\underset{r}{\arg max}{h}_i\[\mathrm{\Ω},r\].$

In other embodiments, estimator f () can comprise map histograms h _i[Ω, r] is to the amplitude calibration factor gradual.This recurrence can use supervised learning technology to train.Such as, user or manufacturer can determine the histogram h for the one group of microphone using similar process to manufacture _i| Ω, r] and amplitude correction factor λ _i[Ω].In some cases, user or manufacturer can use calibrated off-line technology determination histogram h _i[Ω, r] and amplitude correction factor λ _i[Ω].Subsequently, user or manufacturer can determine histogram h _i[Ω, r] and amplitude correction factor λ _iparameter Mapping between [Ω] or nonparametric map.This parameter or nonparametric figure can be considered to estimator f ().This Parameter Mapping can comprise linear function or nonlinear function.Nonparametric function can comprise SVMs, core machine or arest neighbors MM.

In certain embodiments, amplitude calibration module 308 can use maximum likelihood (ML) estimator determination amplitude correction factor histogram h can be maximized by identifying at ML estimator _ithe value of the r of [Ω, r] and estimating

${\widetilde{\mathrm{\λ}}}_{i,T}\[\mathrm{\Ω}\]=\underset{{\mathrm{\λ}}_i\[\mathrm{\Ω}\]}{\arg max}\underset{t}{\Π}p(r_i\[n,\mathrm{\Ω}\]|{\mathrm{\λ}}_i\[\mathrm{\Ω}\]).$

Amplitude calibration module 308 can by following modeling possibility item:

p(r _i[n，Ω]|λ _i[Ω])∝exp(-(r _i[n，Ω]-λ _i[Ω]) ²).

In certain embodiments, the confirmable amplitude correction factor of amplitude calibration module 308 use maximum a posteriori (MAP) estimator.Such as, estimator can identify, for each frequency, and the amplitude calibration factor maximize as follows:

${\widetilde{\mathrm{\λ}}}_{i,T}\[\mathrm{\Ω}\]=\underset{{\mathrm{\λ}}_i\[\mathrm{\Ω}\]}{\arg max}\underset{t}{\Π}p(r_i\[n,\mathrm{\Ω}\]|{\mathrm{\λ}}_i\[\mathrm{\Ω}\])p({\mathrm{\λ}}_i\[\mathrm{\Ω}\]).$

As discussed above, amplitude calibration module 308 can the possibility item of modeling as follows:

p(r _i[n，Ω]|λ _i[Ω])∝exp(-(r _i[n，Ω]-λ _i[Ω]) ²)

In certain embodiments, amplitude calibration module 308 can simulate existing term as smoothly existing, and this is conducive to estimating that the difference of the amplitude calibration factor between side frequency is little.So, the MAP estimator size calibration factor λ that can identify _i[Ω], the possibility improved to greatest extent, keeps amplitude calibration factor lambda simultaneously _i[Ω] smoothness in a frequency domain.In a sense, can the estimation amplitude calibration factor side frequency of low pass filter before level and smooth.Existing a kind of possible smoothing can based on Gaussian Profile, following regulation:

p(λ _i[Ω])∝exp(-α(λ _i[Ω]-λ _i[Ω+ΔΩ]) ²)，α＞0

Wherein Ω+Δ Ω represents the frequency adjoining Ω.Another kind of possible existing smoothly can based on the distribution of other type, such as Laplacian distribution, generalized Gaussian distribution and broad sense laplacian distribution.

In certain embodiments ,/liter value, Γ [Ω], can by solving convex minimization function to determine:

${\widetilde{\mathrm{\λ}}}_{i,T}\[\mathrm{\Ω}\]=\underset{\mathrm{\λ}\[\mathrm{\Ω}\]}{\arg min}\{{\{\mathrm{\λ}\[\mathrm{\Ω}\]-h_{i,\mathrm{\Ω}}^T\left(r\right)\}}^2+\mathrm{\α}|\left|D(\mathrm{\λ}\[\mathrm{\Ω}\])\right|{|}^{\mathrm{\κ}}\}$

Wherein, D is the derivative operator in frequency domain, and a is smoothed intensity.This derivative operator can be first derivative operator, second derivative operator or more in higher derivative operator one.Rule of thumb, L1 regularization (i.e. K=1) effect is fine.This technology is also referred to as total variance.

In certain embodiments, amplitude calibration module 308 can use about microphone statistical simulation existing entry.Such as, supplier can provide the statistics of the distribution of amplitude correction factor λ [Ω] for the microphone sold by supplier.Existing entry can consider that these extra statistical informations about microphone are to estimate the amplitude calibration factor

When amplitude calibration module 308 receives additional TFR sample from data preparation module 304, amplitude calibration module 308 can calculate ratio r based on appended sample _i[n, Ω] also uses the new ratio calculated to reappraise this amplitude calibration coefficient such as, amplitude calibration module 308 can add extra ratio r from time frame T+1 to histogram _i[Ω, n], again amplitude calibration coefficient is estimated with based on the histogram upgraded in this way, along with microphone is along with time detecting adventitious sound signal, its amplitude calibration module 308 can reappraise amplitude correction factor λ _i[Ω], follows the tracks of amplitude calibration factor lambda _iany change of [Ω].

In certain embodiments, amplitude calibration module 308 is by estimating the input audio signal M received on multiple time frame _i[n, Ω] and M _rthe relation determination amplitude calibration factor between the TFR sample of [n, Ω].

Figure 14 illustrates according to some embodiments, determines the method for amplitude correction factor for the relation between the TFR sample by estimating the input audio signal received at multiple time frame.

In step 1402, amplitude calibration module 308 can collect the input audio signal M of multiple time frame _i[n, Ω] and M _rthe TFR sample of [n, Ω].

In step 1404, amplitude calibration module 308 can associate the TFR sample M corresponding to same time frame _i[n, Ω] and M _r[n, Ω].Figure 15 illustrates and to relate to corresponding to same time frame TFR sample M according to some embodiments _i[n, Ω] and M _rthe exemplary scatter diagram of [n, Ω].Each scattering point 1502 on scatter diagram corresponds to the TFR sample M of same time frame _i[n, Ω] and M _rthe value of [n, Ω].

In step 1406, amplitude calibration module 308 can determine the TFR sample M corresponding to same time frame _i[n, Ω] and M _rrelation between [n, Ω].

In certain embodiments, amplitude calibration module 308 can suppose input audio signal M _i[n, Ω] and M _rthe TFR sample of [n, Ω] has linear relationship.Therefore, amplitude calibration module 308 can be configured to determine to describe input acoustical signal M _i[n, Ω] and M _rthe line of the linear relationship between the TFR sample of [n, Ω].

In certain embodiments, amplitude calibration module 308 can be supposed further, represents this TFR sample M _i[n, Ω] and M _rthe line of the linear relationship between [n, Ω] is through the initial point of scatter diagram.Such as, for the TFR sample M shown in Figure 15 _i[n, Ω] and M _r[n, Ω], amplitude calibration module 308 identifiable design TFR sample M _i[n, Ω] and M _rthe linear relationship (with zero offset) that circuit 1504 between [n, Ω] describes.In certain embodiments, amplitude calibration module 308 can use line fitting technique to determine row.Line fitting technique can be designed to identify the line of the total orthogonal distance minimized between scattering Points And lines.Such as, this line fitting technique can be designed to identify the row of the summation of the square quadrature distance minimized between scattering Points And lines.As another example, line fitting technique can be designed, to determine the row of the orthogonal distance specification summation minimized between dispersion Points And lines.

In certain embodiments, amplitude calibration module 308 can suppose input audio signal M _i[n, Ω] and M _rthe TFR sample of [n, Ω] has the relation that arbitrary spline curve can be used to be described.In such embodiments, amplitude calibration module 308 can use spline curve fitting technology identification spline curve.

Phase alignment module 310 can be configured to determine the relative phase-angle error between the i-th microphone and reference microphone two different microphones the phase error that observes the arrival direction θ that the phase delay observed of signal can depend on plane wave and given by the feature of microphone

Fig. 7 illustrates arrival direction θ and the phase error of microphone how to cause the phase difference between the signal that detects.Fig. 7 comprises two microphones, M _r204E and M _i204A, the acoustical signal 702 identical with each microphones.If sound source is away from two microphones, then voice signal can be approximately plane wave 702.Plane wave can the incident line 704 connecting microphone 204 with angle θ 706, is called as arrival direction (DOA).If DOA θ 706 is the integral multiple of π, then plane wave will arrive microphone at one time.In this case, the phase difference between the signal detected by reference microphone and the signal detected by i-th microphone is the relative phase-angle error between reference microphone and i-th microphone function.

But as shown in Figure 7, if DOA θ is not the integral multiple of π, then the phase difference between the signal observed at reference microphone and the signal observed at i-th microphone is relative phase-angle error with the function of DOA θ.In the figure 7, plane wave arrives at angle θ, and wherein plane wave clashed into reference microphone M before hitting i-th microphone _r.In detail in this figure, the plane wave additional distance D that must advance is to arrive i-th microphone.Additional phase error between in the signal of the signal that this additional distance (being the function of DOA θ) makes reference microphone MR observe and i-th microphone observation.Therefore, if DOA θ is not the integral multiple of π, then the phase difference between the signal that the signal observed at reference microphone and i-th microphone observe is relative phase-angle error with the function of this DOA θ.Due to DOA θ, the phase delay between the signal that reference microphone and i-th microphone detect can be represented as η _i[Ω, θ].

Phase delay η _i[Ω, θ], relative phase-angle error and DOA θ is correlated with by the following system of linear equation:

Wherein η _i[Ω, θ] is phase delay, relative phase-angle error, f _sbe sample frequency, Ω is frequency bins, and P represents the quantity (such as, resolution) of the frequency window of STFT, and v is the speed of acoustical signal, r _ibe represent the bivector of i-th microphone relative to the position of this reference microphone, and θ is the DOA of acoustical signal.Phase alignment module 308 is configured to measure the η of described phase delay due to DOA θ _i[Ω, θ], solves relative to two described DOA θ and relative phase-angle error above-mentioned equation, to determine relative phase-angle error

In certain embodiments, linear equation system can solve two steps: for estimating the first step of DOA θ and the relative phase-angle error for determining second step.In some cases, DOA θ can use multiple signal to estimate classification (MUSIC) method.In other cases, DOA θ can use ESPRIT method to estimate.In other cases, DOA θ can use beam-forming method to estimate.

In certain embodiments, the relative phase-angle error of DOA θ can be estimated by the said system of direct solution system of linear equations.Fig. 8 A-8B illustrates according to some embodiments for solving the process of linear equation system.Phase alignment module 310 can use this process to estimate relative phase-angle error suppose phase alignment module 310 not yet receive n=1 before any TFR of acoustical signal.Because phase alignment module 310 does not have about relative phase-angle error or any information of DOA θ, phase alignment module can be the relative phase error of all microphone initialization be that zero (such as, microphone has identical phase characteristic.)

In step 802, phase alignment module 310 can receive the TFR of the voice signal received by described i-th microphone and reference microphone.From the TFR sample received, phase alignment module 310 can measure i-th phase delay between microphone and reference microphone wherein subscript " 1 " represents that this phase delay is relevant to a described TFR sample.Phase delay the TFR value that can associate with reference microphone by comparing the i-th microphone calculates.Particularly, phase delay can calculate as follows:

${\mathrm{\η}}_i^1\[\mathrm{\Ω},\mathrm{\θ}\]=\arg (M_i\[n=1,\mathrm{\Ω}\])-\arg (M_R\[n=1,\mathrm{\Ω}\])$

Wherein arg provides the angle of complex variable.

In step 804, phase alignment module 310 can use measured phase delay solve linear equation system, assuming that relative phase error zero:

$\left[\begin{array}{c}{\mathrm{\η}}_1^1\[\mathrm{\Ω},\mathrm{\θ}\]\\ ...\\ {\mathrm{\η}}_i^1\[\mathrm{\Ω},\mathrm{\θ}\]\end{array}\right]=2\mathrm{\π}\frac{{\mathrm{\Ωf}}_s}{2P}v\left[\begin{array}{c}-r_1-\\ ...\\ -r_i-\end{array}\right]\left[\begin{array}{c}cos{\mathrm{\θ}}^1\\ sin{\mathrm{\θ}}^1\end{array}\right],$

Wherein, θ ¹represent the estimation of DOA at t=1, i > 1.When the quantity of the microphone except reference microphone is 2 (that is, being set to i=2), said system equation can by reversion $2\mathrm{\π}\frac{{\mathrm{\Ωf}}_s}{2P}v\left[\begin{array}{c}-r_1-\\ ...\\ -r_i-\end{array}\right]$ Solve.When the microphone number except reference microphone is greater than 2 (that is, i > 2), then this system was complete, and can use various linear resolver to solve.Such as, phase alignment module 310 can use least squares approach to solve above-mentioned system:

${\mathrm{\θ}}^1=\underset{\mathrm{\θ}}{\arg \min }\left\{{\left\{\left[\begin{array}{c}{\mathrm{\η}}_1^1\[\mathrm{\Ω},\mathrm{\θ}\]\\ ...\\ {\mathrm{\η}}_i^1\[\mathrm{\Ω},\mathrm{\θ}\]\end{array}\right]-2\mathrm{\π}\frac{{\mathrm{\Ωf}}_s}{2P}v\left[\begin{array}{c}-r_1-\\ ...\\ -r_i-\end{array}\right]\left[\begin{array}{c}cos\mathrm{\θ}\\ sin\mathrm{\θ}\end{array}\right]\right\}}^2\right\}$

In step 806, phase alignment module 310 is relevant to solve following equations, be used in the θ that step 804 is measured ¹value and measured phase delay $\left[\begin{array}{c}{\mathrm{\η}}_1^1\[\mathrm{\Ω},\mathrm{\θ}\]\\ ...\\ {\mathrm{\η}}_i^1\[\mathrm{\Ω},\mathrm{\θ}\]\end{array}\right],$ To estimate relative phase-angle error

Step 808-814 presents phase alignment module 310 and how to reappraise relative phase-angle error, when it receives new data sample at n=T.In step 808, phase alignment module 310 receives new sample of signal at n=T, and phase alignment module 310 can measure i-th phase delay between microphone and reference microphone in step 810, phase alignment module 310 can be passed through relative to θ ^tsolve system estimation DOA θ below ^t:

Wherein represent the relative phase-angle error that the data sample using time frame n=T-1 is estimated.In step 812, once estimate DOA θ ^twith T sample, this phase alignment module 310 by about solve following system and estimate interim relative phase-angle error

In certain embodiments, HASE calibration module 310 can regular interim relative phase-angle error adjacent frequency is made to have similar relative phase-angle error.Such as, phase alignment module 310 can by minimize relative to energy function and solve above-mentioned linear system:

Wherein, D is derivative operator in a frequency domain, and a and κ is the parameter controlling normalized amount.This derivative operator can be first derivative operator, second derivative operator or more higher derivative operator.Rule of thumb, L1 regularization (that is, κ=1) effect is fine.

In step 814, phase alignment block 310 can based on interim relative phase-angle error estimated time T relative phase-angle error in certain embodiments, phase alignment block 310 can arrange interim relative phase-angle error relative phase-angle error for time frame T:

In other embodiments, the relative phase-angle error of the renewable time frame T-1 of phasing block 310 use interim relative phase-angle error relative phase-angle error is made significantly to stride across adjacent time frame and acutely to change.Such as, phase alignment block 310 can calculate at the relative phase-angle error of time frame T estimation as follows:

Wherein, be estimated time frame T to frequency omega _prelative phase-angle error; μ represents that more new estimation is at the Learning Step of the relative phase-angle error of time frame T-1; Represent that P-takes advantage of-P transmission matrix with S, μ can owing to controlling based on interim relative phase-angle error and upgrade the renewal rate of relative phase-angle error at T-1.

In some cases, transmission matrix S can be unit matrix.In other cases, transmission matrix can be smoothly at the smoothing operator of the adjacent frequency band of the relative phase-angle error of time frame T-1 estimation.Such as, this transmission matrix can be:

$S=\mathrm{\β}\left[\begin{array}{cccccc}1& 1& 0& .& 0& 0\\ 0& 1& 1& .& 0& 0\\ 0& 0& 1& .& 0& 0\\ 0& 0& 0& .& 0& 0\\ .& .& .& .& .& .\\ 0& 0& 0& .& 1& 1\end{array}\right]+(1-\mathrm{\β})I$

Wherein I is unit matrix, and β controls the degree of former estimation with Frequency Smooth of wherein relative phase-angle error.

The appended sample that step 808-814 can receive in time repeats, as pointed out in step 816.Therefore, phase alignment module 310 can follow the tracks of any change of a period of time of relative phase-angle error.

In certain embodiments, phase correction module 310 optimisation technique of other types can be used interim relative phase-angle error that Combined estimator meets the following system of linear equation and DOA θ:

In certain embodiments, phase correction module 310 can use Gradient Descent optimisation technique, with common relative to this interim relative phase-angle error and direction of arrival θ solves with minor function:

Wherein D is differential operator in a frequency domain, and a and κ is the parameter for controlling normalized amount.The optimized Gradient Descent optimisation technique that can solve the problem can comprise stochastic gradient descent method, conjugate gradient method, Nellie moral-Mead method, Newton method and random element gradient method.In other embodiments, linear equation system can use one mole of Roger Penrose pseudo inverse matrix to solve, disclosed in before.

Fig. 9 A-9C illustrates the progress of amplitude according to some embodiments and phase calibration process.The true alignment profiles in ground represents with point, and estimated calibration curve uses continuous print line to represent.Fig. 9 A illustrates the state of the estimation when correction module 306 is opened at first.Therefore, because calibration module 306 does not receive many data samples, the alignment profiles of estimation is different from the true alignment profiles in ground completely.But when calibration module 306 receives additional data sample in a period of time, as being shown in Fig. 9 B-9C, the alignment profiles of estimation becomes more and more accurate.

In certain embodiments, calibration module 306 can calculate the different calibration curves of the different arrival directions of acoustical signal.In this way, calibration module 306 can compensate the amplitude calibration Summing Factor relative phase-angle error between two microphones more accurately.Accomplish this point, calibration module 306 can use the DOA flag data sample estimated by data preparation module 304, and calculates different calibration curves for each DOA.In certain embodiments, DOA discretely can turn to frequency range.Therefore, calibration module 306 can be configured to the different calibration curves calculating each discrete DOA frequency range, and wherein discrete DOA frequency range can be included in the DOA in preset range.In certain embodiments, calibration module 306 can be configured to the different calibration curves (such as, the approximating 2-3 of its an index frequency range) of the discrete DOA frequency range near calculating.

In certain embodiments, phase correction module 310 can remove the bias voltage because direction relative phase postpones.Such as, phase alignment module 310 can estimate the different relative phase-angle errors of different DOA, and the different relative phase-angle error of equalization is estimated, to determine final relative phase-angle error subsequently.In another example, phase alignment module 310 (1) can select data sample, and being evenly distributed of the DOA be associated with the sample selected is distributed, and (2) only use selected sample to estimate relative phase-angle error.

In certain embodiments, calibration module 306 can from the microphone selection reference microphone of a group (i+1).Theoretically, calibration module 306 can select (i+1) individual microphone any one as reference microphone.But if the reference microphone of Stochastic choice is defective, calibration process may become unstable.In order to address this problem, calibration module 306 can from the suitable reference microphone of (i+1) individual microphone identification.

In certain embodiments, can determine whether should from the microphone of " i " to select new reference microphone for calibration module 306.Such as, if estimate amplitude correction factor value be greater than the lower threshold of regulation or lower than predetermined upper threshold value, calibration module 306 can change reference microphone.In another example, calibration module 306 can safeguard the probabilistic model of expection alignment profiles.If like this, calibration module 306 can use hypothesis testing method to determine that this calibration module 306 should select new reference microphone.In this hypothesis testing method, calibration module 306 can determine alignment profiles as mentioned above.Then, calibration module 306 can determine determined alignment profiles whether according to the probabilistic model of expection alignment profiles.If determine that alignment profiles is not in accordance with probabilistic model, then calibration module 306 can select new reference microphone.

Even if there is multi-acoustical (such as, two people talk mutually) in scene, disclosed calibration module 306 can be healthy and strong.As a rule, the possibility that different sound sources occupies same time-frequency bins [n, Ω] is little.Therefore, TFR sample M _i[n, Ω] would not correspond to multi-acoustical.Even if TFR sample M _i[n, Ω] corresponding to multi-acoustical, when i-th microphone detects other TFR samples corresponding to single sound source, corresponding to the TFR sample M of multi-acoustical _i[n, Ω] on average will fall and can not affect estimation calibration curve in the long run.In some cases, TFR sample M _ithe T/F resolution of [n, Ω] can correspondingly adjust, and the possibility that different sound sources occupies same time-frequency bins [n, Ω] is less.

Once calibration module 306 reappraises the amplitude calibration factor and relative phase-angle error calibration module 306 can provide alignment profiles to data preparation module 304.Then, as discussed as mentioned above, data preparation module 304 can use the alignment profiles reappraised compensate the TFR of input signal and provide it to application module 312.In certain embodiments, calibration module 306 can store calibration curve in memory.

Subsequently, application module 312 can use the data sample of calibration, with enable application program.Such as, application module 312 can be configured to perform acoustical signal blind source separating.Application module 312 also can be configured to perform speech recognition, and to remove background noise from the inlet flow of signal, to improve the audio quality of input signal, or performing beam forming, is specific audio-source to improve the sensitivity of system.Application module 312 can be further configured to and perform 61/764290 and 61/788521 disclosed operation in US provisional patent, and both are entitled as " SIGNALSOURCESEPARATION ", and both is all incorporated herein by reference in their entirety.Such as, application module 312 can be configured to select data sample from specific arrival direction, makes from the acoustical signal of specific direction by the process of block subsequently system.Application module 312 can be configured to perform probability inference.Such as, application module 312 can be configured to perform belief propagation to graphical model.In some cases, described graphical model is based on the graphical model because of sketch map; In other cases, described graphical model can be hierarchy figure model; In other cases, described graphical model can be Markov random field (MRF); In other cases, described graphical model can be condition random field (CRF).

Figure 10 A-10D illustrates the advantage according to the correcting mechanism calibration microphone disclosed in the use of some embodiments.Figure 10 A represents the ground truth arrival direction (DOA) of acoustical signal.The brightness of Figure 10 A represents the radian of DOA.Figure 10 B illustrates the DOA estimated by relative phase-angle error (such as, not having calibration module 306) between uncompensation microphone.Figure 10 C illustrates the DOA estimated by the relative phase error (such as, using calibration module 306) by compensating between microphone.Figure 10 D illustrates the energy of the signal it being estimated to DOA.

In general, compared to the DOA using calibration to estimate, without the more noises of DOA that calibration is estimated.In fact, the DOA without calibration estimates in fact as the function shifts of frequency, and this DOA estimated for calibration is really not so.Therefore, the suggestion calibration of amplitude calibration Summing Factor relative phase-angle error is useful for application module 312.

In addition, in the ordinary course of things, the DOA that calibration is estimated improved along with the time.This phenomenon illustrates, when calibration module 304 receives additional data sample in time, alignment profiles is estimated to become better.When the energy be associated with measured signal lower (such as, the noise level lower than microphone), DOA estimates different stable.This is because when signal level is lower, there is not signal to estimate DOA.In certain embodiments, microphone signal can use the denoising before application module 312 uses of denoising module.

In certain embodiments, calibration module 306 can use auto-adaptive filtering technique to estimate alignment profiles figure 11 illustrates according to the alignment profiles method of estimation at the auto-adaptive filtering technique according to some embodiments.In step 1102, calibration module 306 may be received in the TFR sample of time frame n=T.

In step 1104, calibration module can estimate TFR sample M _ithe DOA θ of [n=T, Ω].As discussed above, in certain embodiments, Multiple Signal Classification (MUSIC) method, ESPRIT method or bundle formation method can be used to estimate DOA θ.

In certain embodiments, by solving the DOA θ of the system estimation input audio signal of linear equation:

$\left[\begin{array}{c}{\mathrm{\η}}_1^T\[\mathrm{\Ω},\mathrm{\θ}\]\\ ...\\ {\mathrm{\η}}_i^T\[\mathrm{\Ω},\mathrm{\θ}\]\end{array}\right]=2\mathrm{\π}\frac{{\mathrm{\Ωf}}_s}{2P}v\left[\begin{array}{c}-r_1-\\ ...\\ -r_i-\end{array}\right]\left[\begin{array}{c}cos\mathrm{\θ}\\ sin\mathrm{\θ}\end{array}\right],$

Wherein, the relative phase delay between i-th microphone and reference microphone (such as, at time frame T), f _sbe the sample frequency at ADC302, Ω is in a frequency domain, and P represents the number of the conversion frequency range (such as, resolution) that T/F is changed, and such as STFT, v are the rotating speeds of acoustical signal, r _ibe represent the bivector of i-th microphone relative to the position of reference microphone, and θ is the DOA of acoustical signal.Linear equation system can solve relative to DOA θ, to find input TFRM _ithe DOA of [n=T, Ω].TFRM _ithe DOA of [n=T, Ω] can be expressed as θ ^t.Relative phase delay about commercial measurement or estimation disclosed in Fig. 4,8 above can using.DOA θ ^tcan use and estimate about technology disclosed in Fig. 4,8 above.

Subsequently, calibration module 306 can compensate TFRM _ithe relative phase delay of [nn=T, Ω], due to DOA θ ^t.This compensation TFR sample can be calculated as follows:

If all microphones have identical amplitude response and identical phase response (such as, being the relative phase-angle error of zero), then the compensation TFR sample of all microphones should be identical.Any difference in TFR sample through compensating can owing to amplitude correction factor and relative phase-angle error.

In a step 1106, calibration module 306 can be changed and compensate TFR sample to time-domain signal such as, calibration module 306 can to compensation TFR sample operations against time-frequency conversion.

In step 1108, calibration module 306 can determine linear filter gi (t), and it maps the time-domain signal of i-th microphone to the time-domain signal of reference microphone

Wherein represent convolution algorithm symbol.Like this, linear filter g _i(t) can consider i-th any relative phase sensitiveness between microphone and reference microphone and relative phase-angle error.Calibration module 306 can calculate the linear filter g of i-th microphone in the microphone array with (i+1) microphone _i(t).

In certain embodiments, the calibration module 306 linear filter g that auto-adaptive filtering technique identification can be used such _i(t).Auto-adaptive filtering technique can comprise lowest mean square filtering technique, the least square recurrence filter technology, many delay blocks adaptive frequency domain filter technology, kernel sef-adapting filter technology and/or dimension and receive Hopf method.Eliminate at acoustic echo the auto-adaptive filtering technique used in application also to can be used for identifying such linear filter g _i(t).

In certain embodiments, alignment profiles can be represented as linear filter g _i(t).In other embodiments, alignment profiles can be represented as linear filter g _ithe TFR of (t).For this reason, in step 1110, calibration module 306 optionally calculates linear filter g _ithe TRF of (t).

In certain embodiments, calibration module 306 can be configured to reduce amount of calculation by interpolation calibration factor in different frequencies.Calibration module 306 can be configured to maintain the amplitude calibration factor and/or the relative phase-angle error of mapping between the amplitude correction factor of (1) one class frequency and/or relative phase-angle error and (2) frequency not included in this class frequency.

During calibration session, calibration module 306 can be configured to amplitude correction factor and/or the relative phase-angle error of determining group of frequencies.Then, not also determine amplitude correction factor and/or the relative phase-angle error of the frequency not included in group of frequencies, calibration module 306 can use the amplitude calibration factor and/or the relative phase-angle error that map and estimate not to be included in the frequency in group of frequencies.In this way, calibration module 306 can reduce the amount of calculation for all frequency determination amplitude calibration factors paid close attention to and/or relative phase-angle error.In some cases, the group of frequencies of the amplitude calibration factor determined of calibration module 306 and/or relative phase-angle error can include as few as a frequency.

In certain embodiments, calibration module 306 can be configured to use regression function to determine to map.In some cases, regression function can be configured to be similar to based on the amplitude calibration factor of group of frequencies and/or relative phase-angle error the one or more parameters not being included in the amplitude correction factor of group of frequencies and/or the spline curve of relative phase-angle error.In other cases, regression function can be configured to estimate the amplitude calibration factor of each frequency not in group of frequencies and/or the actual value of relative phase-angle error based on the amplitude calibration factor of this group of frequencies and/or relative phase-angle error.

Disclosed device and system can comprise computing equipment.Figure 12 is the block diagram of the calculation element according to some embodiments.Block diagram display computing equipment 1200, it comprises processor 1202, memory 1204, one or more interface 1206, data preparation module 304, has the calibration module 306 of amplitude calibration module 308 and phase alignment module 310 and application module 312.Computing equipment 1200 can comprise other suitable combination any of additional module, less module or module, to perform any suitable operation or combination operation.

Equipment 1200 can communicate via interface 1206 with other computing equipment (not shown).Interface 1206 can with hardware implementing, to send and to receive the signal in various medium, such as light, copper and wireless, and in non-transient multiple different agreements some.

In certain embodiments, one or more module 304,306,308,310 and 312 can use memory 1204 to implement in software.Memory 1204 also can keep the calibration curve of microphone.Memory 1204 can be non-temporary computer readable medium, flash memory, disc driver, CD drive, programmable read only memory (PROM), read-only memory (ROM) or other memory any or compound storage.This software can the operation of computer instructions or computer code at processor 1202.Processor 1202 also can use in application-specific integrated circuit (ASIC) (ASIC), programmable logic array (PLA), digital signal processor (DSP), field programmable gate array (FPGA) or other integrated circuit any at hardware and realize.

One or more module 304,306,308,310 and 312 can use ASIC, PLA, DSP, FPGA or other integrated circuit any to realize with hardware.In certain embodiments, two or more module 304,306,308,310 and 312 can be embodied in identical integrated circuit, as ASIC, PLA, DSP or FPGA, thus forms SOC (system on a chip).

In certain embodiments, computing equipment 1200 can comprise subscriber equipment.Subscriber equipment can communicate with one or more radio access network and wireline communication network.Subscriber equipment can be the cell phone with voice communication capability.Subscriber equipment also can be that smart phone provides service, as word processing, web-browsing, game, e-book ability, operating system and full keyboard.Subscriber equipment also can be to provide the flat computer of network insertion and the great majority service provided by smart phone.Subscriber equipment uses operating system (such as, SymbianOS, iPhoneOS, RIM blackberry, blueberry, WindowsMobile, Linux, HPWebOS and Android) operation.Screen can be the touch-screen for entering data into mobile device, and in this case, screen can be used for replacing full keyboard.Subscriber equipment also can keep global positioning coordinates, profile information or other positional informations.

Computing equipment 1200 can also comprise and can calculate and communicate by any platform.Limiting examples can comprise the equipment of computing capability of TV (TV), video frequency projector, Set Top Box or machine top unit, digital video recorder (DVR), computer, net book, notebook computer and any other audio/video.This computing equipment 1200 can be configured one or more processor, the software that its processing instruction and operation store in memory.Processor also uses memory and interface communication, to communicate with other equipment.This processor can be any applicable processor, such as, be combined with the single-chip of CPU, application processor and flash memory system.This computing equipment 1200 can also provide various user interface, such as keyboard, touch-screen, trace ball, touch pads and/or mouse.In certain embodiments, this computing equipment 1200 can also comprise loud speaker and display device.

Computing equipment 1200 can also comprise biomedical electronic equipment.Biologic medical electronic equipment can comprise hearing aids.This computing equipment 1200 can be consumer device (such as, television set or microwave oven), and calibration module can promote that Voice command is carried out in the audio frequency input strengthened.In certain embodiments, computing equipment 1200 can be integrated into larger system, so that audio frequency process.Such as, computing equipment 1200 can be a part for automobile, and can promote everybody and/or man-machine communication.

Figure 13 A-13B illustrates the one group of microphone that can be used for the calibration process disclosed in combination according to some embodiments.This group microphone can be placed on microphone unit 1302.Microphone unit 1302 can comprise multiple microphone 204.Each microphone can comprise MEMS element 1306 and be coupled to one of four ports arranged at 1.5 millimeters of-2 millimeters of square structures.The MEMS element of multiple microphone can share public rear volume 1304.Selectively, each element can by volume after independent distribution.

More generally, microphone comprises multiple port, and multiple element is connected respectively to one or more port, and may be coupled between port (such as, with coupling specific between port or use one or more public rear volume).This more complicated arrangement can eliminate characteristic to provide suitable input, to process further in conjunction with physical orientation, frequency and/or noise.

In certain embodiments, microphone unit 1302 also can comprise one or more data preparation module 304, amplitude calibration module 308 and phase alignment module 310.By this way, microphone unit 1302 can become self calibration microphone unit, and it can be coupled to computing system, and does not need computing system to calibrate the voice data of microphone unit 1302.In some cases, the phase alignment module 310 in data preparation module 304, amplitude calibration module 308 and/or microphone unit 1302 may be implemented as hard-wired system.In other cases, the phase alignment module 310 in data preparation module 304, amplitude calibration module 308 and microphone unit 1302 can be configured to make processor perform the method step relevant to modules.In some cases, microphone unit 1302 also can comprise application module 312, thus provides intelligent microphone unit.

Microphone unit 1302 can use interface to communicate with miscellaneous equipment.Interface can realize within hardware, to send and to receive the signal in various medium, and such as light, copper and wireless, and wherein non-transient many different agreement.

Should be appreciated that disclosed theme is not limited to the details being applied to structure, and the layout of parts shown in the description be shown in below or accompanying drawing.Disclosed theme by practice and can carry out in every way.In addition, be appreciated that the wording that this paper adopts and term are the objects in order to describe, and should not be considered to restriction.

Like this, it will be apparent to one skilled in the art that the disclosure based on concept easily can be used as designing other structure, method and system basis, some objects of the open theme of application implementation.This point is very important, and therefore, claim is believed to comprise these equivalent constructions, as long as they do not depart from the spirit and scope of disclosed theme.Such as, disclosed step can be undertaken by one or more variable.This relation can represent with mathematical formulae.But those of ordinary skill in the art also can use different mathematical equations to express the relation between one or more variable by the mathematical equation disclosed in conversion.Importantly, claim is regarded as such equivalent relation of comprising between one or more variable.

Although describe at preceding example and embodiment theme of the present disclosure be shown, it should be understood that, present disclosure carries out by means of only the mode of citing, and can carry out many changes to the details of disclosed theme, and does not deviate from the spirit and scope of disclosed theme.

Claims (40)

1. a device, comprising:

Interface, is configured as receiving the first digitized signal stream and the second digitized signal stream, and wherein, described first digitized signal stream and the second digitized signal stream correspond to the voice signal of being caught by the first microphone and second microphone respectively;

With the processor of described interface communication, be configured for operation in the module stored in memory, wherein said module is configured to:

Determine the first digitized signal stream the very first time-the second T/F of frequency and the second digitized signal stream represents, wherein, the very first time-frequency representation indicates the first digitized signal stream in the amplitude of multiple time frame for multiple frequency, and wherein said second T/F represents that instruction second digitized signal stream is in the amplitude of multiple time frame for multiple frequency;

Determine the very first time-frequency representation and the second T/F represent at multiple time frame for the relation between more than first frequency; With

Based on the very first time-frequency representation and the second T/F represent, determine the first microphone and second microphone for more than first frequency the amplitude calibration factor between relation.

2. device as claimed in claim 1, wherein, described module be configured to determine described the very first time-frequency representation and by the relation between the second T/F representation:

For more than first frequency, determine described second T/F represent with the very first time-frequency representation is for the ratio of each described multiple time frame; With

Determine the histogram of the ratio corresponding to multiple frequency described in first.

3. device as claimed in claim 2, wherein, described module is configured to determine the amplitude calibration factor based on the counting of ratio in histogram.

4. device as claimed in claim 3, wherein, described module is further configured to:

Determine based on multiple histogram the multiple amplitude calibration factors corresponding to multiple frequency, wherein said multiple histogram corresponds respectively to multiple frequency; With

The amplitude calibration factor that level and smooth and at least two described multiple frequency dependences join.

5. device as claimed in claim 3, wherein, described module is configured to the ratio by identifying the highest counting in histogram, and determines the amplitude calibration factor of more than first frequency.

6. device as claimed in claim 1, wherein, described module is configured to by identifying line and determine relation, described line modeling the very first time-frequency representation and the second T/F represent corresponding to the relation between described multiple time frame and described more than first frequency.

7. device as claimed in claim 1, wherein, described module be configured to be multiplied by more than first frequency described the very first time-amplitude correction factor of frequency representation and more than first frequency, to calibrate the first microphone relative to second microphone.

8. device as claimed in claim 1, wherein, described module is further configured to:

Receive and correspond to by the first additional character signal of the first microphone at the first digitized signal stream of the voice signal of very first time frame-grab;

Receive and correspond to by the second additional character signal of second microphone at the second digitized signal stream of the voice signal of very first time frame-grab;

Based on the described first additional digitized signal, calculate the 3rd time-frequency representation;

Based on the signal of described second additional character, calculate the 4th time-frequency representation; With

Represent and described 4th time-frequency representation based on the 3rd T/F, upgrade amplitude calibration factor.

9. device as claimed in claim 8, wherein, described module is configured to:

The amplitude that the 3rd T/F being identified in very first time frame represents lower than the frequency of noise level, and

When upgrading amplitude correction factor based on the 3rd time-frequency representation, abandon the 3rd time-frequency representation of identified frequency and very first time frame.

10. device as claimed in claim 8, wherein, described module is configured to:

The 3rd T/F being identified in very first time frame represents the frequency relevant to undesirable acoustical signal;

When upgrading amplitude correction factor based on the 3rd time-frequency representation, abandon the 3rd time-frequency representation of identified frequency and very first time frame.

11. devices as claimed in claim 10, wherein, when the ratio of the 4th time-frequency representation and described 3rd time-frequency representation be enough different from based on the very first time-frequency representation and the second T/F represent that the amplitude calibration of calculating is because of the period of the day from 11 p.m. to 1 a.m, described module is configured to determine that the 3rd time-frequency representation is associated with undesirable acoustical signal.

12. devices as claimed in claim 1, wherein, described time-frequency representation comprises one or more short time discrete Fourier transforms (STFT) or wavelet transformation.

13. 1 kinds of methods, comprising:

The first digitized signal stream and the second digitized signal stream is received by the data processing module being coupled to the first microphone and second microphone, wherein, described first digitized signal stream and the second digitized signal stream correspond to the voice signal of catching respectively by the first microphone and second microphone;

By data processing module determine described first digitized signal stream the very first time-the second T/F of frequency representation and the second digitized signal stream represents, wherein, the very first time-frequency representation represents the size of the first digitized signal stream for multiple frequencies of multiple time frame, and wherein said second T/F expression second digitized signal stream is for the amplitude of multiple frequencies of multiple time frame;

By the calibration module communicated with described data processing module determine described the very first time-frequency representation and the second T/F represent between in the relation of multiple time frame for multiple first frequency;

Based on the very first time-relation between frequency representation and the second T/F representation, determine the first microphone and the second microphone amplitude calibration factor for more than first frequency by calibration module.

14. methods as claimed in claim 13, wherein, determine described the very first time-relation between frequency representation and the second T/F representation comprises:

For more than first frequency, determine described second T/F represent with the very first time-frequency representation is for the ratio of each described multiple time frame; With

Determine the histogram of the ratio corresponding to multiple frequency described in first.

15. methods as claimed in claim 13, wherein, determine described the very first time-relation between frequency representation and the second T/F representation comprises identification line, described line modeling the very first time-frequency representation and the second T/F represent corresponding to the relation between described multiple time frame and described more than first frequency.

16. methods as claimed in claim 13, comprise further: be multiplied by more than first frequency described the very first time-amplitude correction factor of frequency representation and more than first frequency, to calibrate the first microphone relative to second microphone.

17. methods as claimed in claim 13, comprise further:

Receive and correspond to by the first additional character signal of the first microphone at the first digitized signal stream of the voice signal of very first time frame-grab;

Receive and correspond to by the second additional character signal of second microphone at the second digitized signal stream of the voice signal of very first time frame-grab;

Represent based on described first additional character calculated signals the 3rd T/F;

Represent based on described second additional character calculated signals the 4th T/F;

To represent with the 4th T/F upgrade the amplitude calibration factor with representing based on the 3rd T/F.

18. 1 kinds of non-transitory computer-readable mediums, have and can operate to make the executable instruction of data processing equipment:

By being coupled to interface first digitized signal stream and the second digitized signal stream of the first microphone and second microphone, wherein said first digitized signal stream and the second digitized signal stream correspond to the voice signal of being caught by described first microphone and second microphone respectively;

Determine described first digitized signal stream the very first time-the second T/F of frequency representation and the second digitized signal stream represents, wherein, the very first time-frequency representation indicates the first digitized signal stream for the size of multiple frequencies of multiple time frame, and wherein said second T/F represents the size of instruction second digitized signal stream for the multiple frequencies for multiple time frame;

Determine described the very first time-frequency representation and described second T/F representation at multiple time frame for the relation between more than first frequency, and

Based on the very first time-relation between frequency representation and the second T/F representation, determine the first microphone and the second microphone amplitude calibration factor for more than first frequency.

19. non-transitory computer-readable mediums as claimed in claim 18, wherein, described executable instruction can operate to make data processing equipment identification line, described line modeling the very first time-frequency representation and the second T/F represent corresponding to the relation between described multiple time frame and described more than first frequency.

20. non-transitory computer-readable mediums as claimed in claim 18, wherein said executable instruction can operate to make data processing equipment with:

Receive and correspond to by the first additional character signal of the first microphone at the first digitized signal stream of the voice signal of very first time frame-grab;

Receive and correspond to by the second additional character signal of second microphone at the second digitized signal stream of the voice signal of very first time frame-grab;

Represent based on described first additional character calculated signals the 3rd T/F;

Represent based on described second additional character calculated signals the 4th T/F; With

Represent based on the 3rd T/F and to represent with the 4th T/F and upgrade the amplitude calibration factor.

21. 1 kinds of devices, comprising:

Be configured to the interface of reception first digitized signal stream and the second digitized signal stream, wherein, described first digitized signal stream and the second digitized signal stream correspond to the voice signal of being caught respectively by the first microphone and second microphone;

Processor, with described interface communication, be configured to the module stored in run memory, wherein said module is configured to:

Determine described first digitized signal stream the very first time-the second T/F of frequency representation and the second digitized signal stream represents, wherein, the very first time-frenquency representation indicates the first digitized signal stream relative to the amplitude of multiple frequencies of very first time frame, and described second T/F representation represents the amplitude of the second digitized signal stream relative to described multiple frequency and very first time frame;

Positioned opposite based on described first microphone and second microphone, and more than first frequency and very first time frame the very first time-frequency representation and the second T/F represent and calculate the first parameter of the arrival direction of instruction voice signal;

Based on the first parameter, the very first time-frequency representation and the second time-frequency representation at more than first frequency and very first time frame, determine the first relative phase-angle error for very first time frame and more than first frequency between described first microphone and second microphone.

22. devices as claimed in claim 21, wherein, described module is configured to:

Determine the very first time-frequency representation and the second T/F representation be in the first-phase potential difference of described more than first sampling frequency between very first time frame; With

Based on first-phase potential difference, determine the first parameter.

23. devices as claimed in claim 21, wherein, described module is further configured to determines the first parameter according to linear system, described linear system relate at least partly arrival direction and the very first time-phase difference between frequency representation and the second T/F representation.

24. devices as claimed in claim 21, wherein, described module is further configured to:

Receive the first additional character signal of the first digitized signal stream corresponding to the voice signal of being caught at the second time frame by the first microphone;

Receive the second additional character signal of the second digitized signal stream corresponding to the voice signal of being caught at the second time frame by second microphone;

Based on described first additional character signal, the 3rd T/F calculating the second time frame represents;

Based on described second additional character signal, the 4th T/F calculating the second time frame represents;

Based on the first relative phase-angle error of positioned opposite, the very first time frame for the 3rd frequency representation of the second time frame and the 4th frequency representation, the first microphone and second microphone, determine to represent second parameter of described acoustical signal for the arrival direction of the second time frame; With

Based on the 3rd frequency representation and the 4th frequency representation in the second time frame and the second parameter, determine the second relative phase-angle error for the second time frame and more than first frequency between the first microphone and second microphone.

25. devices as claimed in claim 24, wherein, described module is configured to determine the second relative phase-angle error based on described first relative phase-angle error, with relative to described first relative phase-angle error smoothly described second relative phase-angle error.

26. devices as claimed in claim 24, wherein, described module be configured to determine when the first parameter and the second parameter closer to each other time the second relative phase-angle error time, the discretization of the arrival direction of described first parameter instruction very first time frame, and the second parameter indicates the discretization of the arrival direction of the second time frame.

27. devices as claimed in claim 24, wherein, described module is configured to provide mask, and the amplitude of described mask identification the 3rd time-frequency representation is less than the frequency of noise level.

28. devices as claimed in claim 27, wherein, described module is configured to use described mask, with abandon at estimation second relative phase-angle error the 3rd time-frequency representation of identification frequency.

29. devices as claimed in claim 24, wherein, described module is configured to provide mask, and this mask is identified in the 3rd time-frequency representation frequency relevant to undesirable acoustical signal.

30. devices as claimed in claim 29, wherein, described module is configured to use described mask, with abandon in estimation second relative phase-angle error the 3rd time-frequency representation of identification frequency.

31. devices as claimed in claim 21, wherein, described module is configured to the first relative phase-angle error be associated that is level and smooth and at least two described multiple frequencies.

32. methods as claimed in claim 21, wherein said module is configured to:

Receive the first additional character signal of the first digitized signal stream corresponding to the voice signal of being caught at the second time frame by the first microphone;

The 3rd T/F calculating the second time frame based on described first additional signal digitlization represents; With

The first relative phase-angle error is removed, to calibrate the first microphone for frequency multiple described in first relative to described second microphone from the 3rd time-frequency representation for more than first frequency and the second time frame.

33. 1 kinds of methods, comprising:

The first digitized signal stream and the second digitized signal stream is received by the data processing module being coupled to the first microphone and second microphone, wherein, described first digitized signal stream and the second digitized signal stream correspond respectively to the voice signal of being caught by the first microphone and second microphone;

Data processing module determine described first digitized signal stream the very first time-the second T/F of frequency representation and the second digitized signal stream represents, wherein, the very first time-frequency representation indicates the first digitized signal stream relative to the phase place of multiple frequency and very first time frame, and wherein said second T/F representation represents the phase place of the second digitized signal stream for multiple frequency and very first time frame;

The first parameter is calculated at the calibration module communicated with data processing module, its based on described first microphone and second microphone positioned opposite, described in first multiple frequency and very first time frame the very first time-frequency representation and the second T/F represent, the arrival direction of instruction voice signal; With

At calibration module, based on described first parameter, the very first time-frequency representation and representing at the second T/F of more than first frequency and very first time frame, determine the first microphone and second microphone the first relative phase-angle error for more than first frequency and very first time frame.

34. methods as claimed in claim 33, wherein, calculate described first parameter and comprise:

Determine the very first time-frequency representation and the second T/F represent for the first-phase potential difference of more than first sampling frequency between very first time frame; With

Based on first-phase potential difference, determine the first parameter.

35. methods as claimed in claim 34, wherein, determine the first parameter based on first-phase potential difference: comprise and determine the first parameter according to linear system: described linear system relate at least partly arrival direction and the very first time-phase difference between frequency representation and the second T/F representation.

36. methods as claimed in claim 33, also comprise:

Receive the first additional character signal of the first digitized signal stream corresponding to the voice signal of being caught at the second time frame by the first microphone;

Receive the second additional character signal of the second digitized signal stream corresponding to the voice signal of being caught at the second time frame by second microphone;

Based on described first additional character signal, the 3rd T/F calculating the second time frame represents;

Based on described second additional character signal, the 4th T/F calculating the second time frame represents;

Based on the first relative phase-angle error of positioned opposite, the very first time frame for the 3rd frequency representation of the second time frame and the 4th frequency representation, the first microphone and second microphone, determine to represent second parameter of described acoustical signal for the arrival direction of the second time frame; With

Based on the 3rd frequency representation and the 4th frequency representation in the second time frame and the second parameter, determine that the first microphone and second microphone are for the second relative phase-angle error between the second time frame and more than first frequency.

37. methods as claimed in claim 36, wherein, determine that described second relative phase-angle error comprises: determine the second relative phase-angle error based on the first relative phase-angle error, with relative to level and smooth second relative phase-angle error of described first relative phase-angle error.

38. 1 kinds of non-transitory computer-readable mediums, have and can operate to make the executable instruction of data processing equipment, with:

The first digitized signal stream and the second digitized signal stream is received by the data processing module being coupled to the first microphone and second microphone, wherein, described first digitized signal stream and the second digitized signal stream correspond respectively to the voice signal of being caught by the first microphone and second microphone;

Determine described first digitized signal stream the very first time-the second T/F of frequency representation and the second digitized signal stream represents, wherein, the very first time-frequency representation indicates the first digitized signal stream relative to the phase place of multiple frequency and very first time frame, and wherein said second T/F representation represents the phase place of the second digitized signal stream for multiple frequency and very first time frame;

Calculate the first parameter, its based on described first microphone and second microphone positioned opposite, described in first multiple frequency and very first time frame the very first time-frequency representation and the second T/F represent, the arrival direction of instruction voice signal; With

Based on described first parameter, the very first time-frequency representation and representing at the second T/F of more than first frequency and very first time frame, determine the first microphone and second microphone the first relative phase-angle error for more than first frequency and very first time frame.

39. non-transitory computer-readable mediums as claimed in claim 38, wherein said executable instruction can operate to make data processing equipment with:

Determine the very first time-frequency representation and the second T/F represent in described more than first sampling frequency and the first-phase potential difference between very first time frame; With

The first parameter is determined based on first-phase potential difference.

40. non-transitory computer-readable mediums as claimed in claim 38, wherein said executable instruction can operate to make data processing equipment with:

Receive the first additional character signal of the first digitized signal stream corresponding to the voice signal of being caught at the second time frame by the first microphone;

Receive the second additional character signal of the second digitized signal stream corresponding to the voice signal of being caught at the second time frame by second microphone;

Based on described first additional character signal, the 3rd T/F calculating the second time frame represents;

Based on described second additional character signal, the 4th T/F calculating the second time frame represents;

Based on the first relative phase-angle error of positioned opposite, the very first time frame for the 3rd frequency representation of the second time frame and the 4th frequency representation, the first microphone and second microphone, determine to represent second parameter of described acoustical signal for the arrival direction of the second time frame; With

Based on the 3rd frequency representation and the 4th frequency representation in the second time frame and the second parameter, determine that the first microphone and second microphone are for the second relative phase-angle error between the second time frame and more than first frequency.